Comedians test ChatGPT’s humor. Are we laughing with AI or at it?

SAN FRANCISCO — “How many tickles does it take to make an octopus laugh?” Angel Jin, a 27-year-old tech worker and stand-up comedian, asks the audience.

“10 tickles?,” she offers.

Jin isn’t fazed. At this comedy show, it’s good when certain punchlines flop. It was a test: Could the audience guess which jokes the comedian dreamed up and which were written by the popular artificial-intelligence tool ChatGPT?

Everyone’s talking about AI’s potential to kill us, but so far it can’t even kill on the comedy stage.

On a recent Saturday night, Jin and several other amateur comedians performed short stand-up sets and then delivered four one-liners, inviting the crowd to judge which jokes sprang from a human’s brain and which were robot-generated. ChatGPT mostly turned up dad jokes lifted from the internet, making it easy to identify a generic punchline from an original.

“You’re hitting on that sore spot” of ChatGPT potentially taking people’s jobs, “but laughing at it because AI is still so bad,” says comedian Geulah Finman, 31. The show “felt like a release.”

Tests of AI humor — which are being replicated in other comedy clubs and by researchers — are key to helping better understand the technology, as well as the potential risks it poses to us. Experts say that one of the major dangers of AI is its potential to better imitate humans and replicate them, from emotional responses to telling jokes.

While voice assistants like Siri and Alexa have long spouted punchlines, those are preprogrammed and non-interactive. ChatGPT and other bots have the ability to scrape the internet and potentially come up with their own creative versions.

One such joke turned up in a research paper this month by two German researchers. “Why did the man put his money in the blender? He wanted to make time fly.” It was nonsensical, though it showed some creative flair.

But more than 90 percent of the 1,000-plus jokes ChatGPT spit back in the experiment were the same 25 jokes, most of them built on wordplay and puns. Fittingly, two of the common jokes the researchers highlighted — “Why did the tomato turn red? Because it saw salad dressing” and “Why don’t scientists trust atoms? Because they make up everything” — were also in the rotation at the San Francisco show.

“ChatGPT has not solved computational humor yet but it can be a big leap toward ‘funny’ machines,” Darmstadt University researchers Sophie Jentzsch and Kristian Kersting wrote in their paper.

Companies that specialize in the cutting-edge technology are seeing their stock prices soar as demand for their products skyrockets. City leaders in San Francisco, where many AI start-ups are based, are hoping that the AI gold rush will revive the local tech scene.

A Washington Post analysis said one snapshot showed 15 million websites have informed some high-profile English-language AIs. Models like ChatGPT are helping software engineers create computer code and can even pass the bar exam. But as Hollywood writers strike over the potential for the technology to disrupt their jobs, the demonstrations and research imply the technology could take awhile to catch up.

Naomi Fitter, an assistant professor of robotics at Oregon State University, studies how robots might assist humans in health contexts, like guiding people through physical therapy exercises. Starting in 2018, Fitter wrote stand-up comedy routines for a robot she named Jon and sent him out on tour in Los Angeles.

Jon the Robot uses artificial intelligence to determine where to jump next in his human-written script. Jon can tell a joke has fallen flat, Fitter says, and then make a quip on the joke’s failure, attempting to repair the interaction. “It might be poking fun at the audience, trying to guess why they didn’t like the joke,” Fitter says. The majority of the time, when the robot tried to rescue the joke, it improved the audience’s reaction, a result Fitter finds “promising.”

“You have been a great audience,” Jon tells an audience in a 2020 YouTube video. “If you like me, please book me and help me take your jobs.”

Humor generally requires a careful combination of the mundane and the absurd — and so far, ChatGPT lacks the brevity and creativity to be funny, comedy experts say. Except, of course, when it “hallucinates,” or volunteers inaccurate information.

“The humor comes from how bad the AI is,” says Victor Treviño, a 33-year-old engineer and stand-up comedian who splits his time between San Francisco and Los Angeles. Treviño produces a show where comedians do stand-up sets while playing with an AI image generator onstage.

“It’s funny to see what the AI image generator will assume about someone,” like placing them in a scene from the 1960s or giving them extra hands, he says. “It’s like a playground for me.”

When David Isaacs, chair of the division of writing for screen and TV at the University of Southern California’s School of Cinematic Arts, asked ChatGPT to write a movie scene where a man is having trouble telling a woman he loves her, the program spit out three pages “without much flair or without really a curve on it,” Isaacs said. He noted it could be a way out of writer’s block.

Still, “it gets me somewhere,” he added. “It took me out of the tyranny of the open page.”

Some comedy writers see a need for artificial intelligence to master the art of levity. Years ago, while reading about the loneliness epidemic, Joe Toplyn, a former writer for sitcoms and late-night TV, figured that eventually people might become more accepting of artificial companions — and those companions would need a sense of humor.

Toplyn, who has degrees in engineering and applied physics, has used the AI tools to build a joke chatbot he named Witscript. Imagine a more concise, slightly absurd version of ChatGPT. Like Isaacs, Toplyn sees potential in the misfires.

“It might give you an idea for another joke if Witscript turns out a joke that’s not quite there,” he says.

AI is a frequent topic of conversation in the Bay Area — so naturally it’s made its way to the comedy stage.

Stroy Moyd, a 35-year-old comedian, got the idea for an AI-themed comedy night dubbed LaughGPT after overhearing audience members at another show talking about the hyped technology. LaughGPT sold out faster and with less effort than Moyd usually puts into marketing, he said.

“It was just an experiment,” Moyd said before the evening’s recent back-to-back shows, which drew tech enthusiasts from their 20s to their late 60s.

When it comes time to test the ChatGPT punchlines in San Francisco, comedian Finman lapses into a generic joke-telling voice — a little nasally and wobbly — priming the audience that a joke isn’t original.

“What’s the deal with airplane food?,” Finman said. “The flavors are so plain. And the prices are sky-high.”

Toward the end of the show, the audience is finally stumped.

“My girlfriend broke up with me for making too many Linkin Park references,” comedian Josef Anolin, 42, says as he wraps his set. “But in the end, it doesn’t even matter.”

“You!” the audience yells.

“That was ChatGPT, baby!”

The crowd roars louder than it has all night.

Leave a Reply

Your email address will not be published. Required fields are marked *